Watermarks are coming to an AI model near you
Creating, publishing, and distributing disinformation of all sorts has never been easier than it is today. Arguably, visual generative AI platforms have brought it to a whole new level: where before the ability of an average person would be limited by editing existing images or footage, it is now possible to generate a completely new thing by simply entering a textual prompt.
We've all been trying to fully grasp the possible consequences of this tectonic shift over the past months, but the realm of possibilities seems to be nearly endless. That's why governments across the world are increasingly racing to regulate (generative) AI applications, as well as create sovereign computing platforms to train their own models.
Regulation, however, is usually unwieldy and takes time to take effect. That leaves the question for everyday people consuming content online: how do we make sure that we know if we're looking at an AI-generated image or video?
One obvious answer to that is reliable watermarking — which turns out to be possible, at least from the technical standpoint. Google has just announced its SynthID system that introduces an invisible watermark that can be detected by an algorithm even after modifications and cropping of the original image.
At this point, of course, there are more questions than answers about the whole idea. Should we mandate all platforms to introduce a watermark like this? Then how about open-source solutions where anyone can modify the code and create watermark-free assets?
Would it maybe make sense to watermark images that are?not?AI-generated? Say, have all camera manufacturers add this same kind of watermark at the time a photo is taken? Most probably there's already a committee working on a draft proposal somewhere deep down the corridors of the European Commission.
In the meantime, check out Thomas Macaulay's write-up on SynthID and the authorities' push to control the generative AI chaos ??
?
领英推荐
What else we’re writing about
???? Myanmar genocide refugees are?taking Meta to Ireland’s High Court?for the tech giant’s alleged role in the 2017 Myanmar genocide, one of the worst war crimes of this century. In 2018, UN human rights investigators said the use of Meta’s social media platform Facebook had played a key role in spreading hate speech that fuelled the violence.
?? A new discovery could lead to?faster-charging and longer-range EVs. In a breakthrough for the EV industry, European researchers have discovered a way to prevent a damaging and potentially dangerous side effect that can occur during fast charging: lithium plating.
??? After nearly 2,000 years since Vesuvius’ eruption buried the vibrant Roman city of Pompeii under a deadly blanket of ash,?we now know a bit more?about the dramatic event. Applying a new research and technological method, a European team led by the University of Valencia has determined that a number of the fossilised Pompeians died of asphyxiation — and not from dehydration or heat as previously theorised.
??? Two Nordic companies will provide the battery and propulsion technology for?the world’s largest electric ship, as the industry looks to charter a course toward greener voyages. The 130-metre-long battery-electric vessel is being built on the island of Tasmania, off the coast of Australia.